After migrations run, any sessions made with the labs flag turned off
will have the verified flag set. We also need new sessions made after
that to gain the verified flag, so that they aren't logged out at the
point that the labs flag is enabled (or removed).
ref ENG-1641
Using `getUserFromSession` requires the cookie header to be set, but
at this point we may still be constructing the session. Instead we can
get the user id from the session itself
More typical in TOTP setups for each token to last 1 minute, and to
allow some older tokens.
Also moved the options setting out of the generate scope in case
verify is called first (unlikely but possible).
closes https://linear.app/tryghost/issue/ENG-1617/
closes https://linear.app/tryghost/issue/ENG-1619/
- updated cookie authenticator's `authenticate` method to accept an `{identification, pasword, token}` object
- if `token` is provided, hit our `PUT /session/verify/` endpoint passing through the token instead of hitting the `POST /session/` endpoint
- added `signin/verify` route
- displays a 2fa code input field, including required attributes for macOS auto-fill from email/messages to work
- uses `session.authenticate({token})` when submitted
- updated signin routine to detect token-required state
- detects a `403` response with a `2FA_TOKEN_REQUIRED` code property when authenticating
- if detected transitions to the `signin/verify` route
refs ENG-1622
Currently unused by the API, this session variable will be used to
confirm whether the user has authenticated their session with an email
OTP. The verified status is not removed on logout, so sessions are now
retained instead of being destroyed.
fixes
https://linear.app/tryghost/issue/DES-483/improve-analytics-card-layout-in-settings
Analytics settings in admin were difficult to read, as they were all
bunched together. This change lays them out in rows, making the
Analytics settings much easier to scan.
---------
Co-authored-by: Steve Larson <9larsons@gmail.com>
Co-authored-by: Kevin Ansfield <kevin@lookingsideways.co.uk>
Co-authored-by: Cathy Sarisky <42299862+cathysarisky@users.noreply.github.com>
Co-authored-by: Ghost CI <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: jubi-git <117194340+jubi-git@users.noreply.github.com>
Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>
ref 324211f
- this includes changes to improve package size
Package size was found to be bloated due to expanding i18n strings. We
were packing all i18n strings instead of just the ones relevant to the
package. Thanks to @cathysarisky for identifying this!
no ref
This change to vite.config.js fixes errors in the sodo-search,
signup-form, and comments packages, preventing inclusion of all of
i18n/locales. As the number of translated strings has gotten larger,
these builds have increased significantly in size, bloated with strings
that aren't actually present in them (in 58 different languages!)
No obvious build errors are present with these changes, but an extra
eyes would be appreciated. Vite isn't my forte.
---------
Co-authored-by: Steve Larson <9larsons@gmail.com>
ref https://linear.app/tryghost/issue/ENG-1653
- we were always setting a `style="background-color: #123456"` attribute on the buttons but that didn't allow for different button states such as the red failure state to correctly override meaning there was some odd behaviour when hovering
- removed the fixed `style` attribute and adjusted `<GhTaskButton>`
- added `@useAccentColor` prop
- when `@useAccentColor` is true, add the necessary `style` attribute except when showing the failure state
fixes
https://linear.app/tryghost/issue/DES-73/enter-should-create-new-navigation-itemmove-to-next-field
Solves a regression where, when adding navigation items, pressing ENTER
(when on the last item in the list) no longer created a new row.
Also solves an issue where the `+` button next to the row of input
fields wasn't positioned correctly when a form error was showing (it
jumped down along with the error).
no issue
- The `test:*` commands in `ghost/core` are all implicitly dependent on
the TS packages in the whole monorepo being built, but we hadn't
explicitly declared this dependency to NX.
- Now if you run `yarn nx run ghost:test:e2e` (or any other `test:*`
commands in ghost), NX will know that it needs to rebuild the TS
packages, unless they are cached and haven't changed.
- With this, you should be able to directly clone the repo and run `yarn
nx run ghost:test:e2e` to run e2e tests, without running `yarn dev` or
`yarn nx run-many -t build:ts` first.
- This is especially useful for getting tests to run properly in docker
no issue
- I apparently never added @tryghost/metrics-server as a dependency to
ghost/core/package.json. It worked in most cases as a 'phantom
dependency' — yarn installs all node_modules in a flat structure, so
even though it wasn't a dependency in package.json, it still resolved to
the correct package, as long as the typescript packages were all built
first.
- This passed CI because we explicitly run ts:build on all packages
before running tests, and it worked in production because we build the
TS packages as part of the docker build. However, when trying to run
tests locally, it would sometimes fail unless you explicitly ran nx
run-many -t build:ts at the top level before running the tests.
- Adding it as a dependency in package.json fixes this problem.
no issue
- Running `yarn test:unit` fails unless you've previously run `yarn
build:ts`. This change tells `nx` about this dependency, so it will run
`yarn build:ts` before `yarn test:unit`. It should take advantage of the
nx cache though, so it will only run the `build:ts` task if necessary
ref https://linear.app/tryghost/issue/ONC-387/
With some recent changes, we added validation to unsubscribe URLs to verify the source, allowing us to cut down on spam and improving security, as the underlying key could be re-generated should the need arise. This had the side effect of making unsubscribe URLs difficult to reconstruct when using third-party/downstream integrations, such as ActiveCampaign, which fills a gap in the current Ghost feature set.
Now any authenticated query to `/api/members` will return an `unsubscribe_url` field that can be used directly.
no ref
When running tests, occasionally we'll see some varying sort in the
members api response because members are generally all created with the
same timestamp. While `ObjectId` should be progressive, and our defalut
sort is `ORDER BY created_at desc, id desc`, we still would sometimes
see issues. This ought to remove any flakiness.
no ref
Stubbed expected test errors. In general, we should be expecting these
errors in the tests as we write them as that is the expected behavior
(or that behavior should change).
closes#16748
The members/:member_id/signin_urls endpoint currently only does
cookie-based authentication. When #21249 is merged, turning on 2FA is
going to break any 3rd party processes that use it (including my social
sign-in offering).
This patch gives admin API keys 'read' permission on this endpoint, and
enables 3rd party processes to handle user logins the right way, instead
of via a staff member's email/password.
Migration included. Feedback appreciated.
I have the wrong name on my migration. I can see it doesn't follow the
naming convention, but I'm not sure how the names are generated.
---------
Co-authored-by: Michael Barrett <mike182uk@gmail.com>
